The role of extracellular matrix in mouse and human corneal neovascularization

Corneal neo-vascularization (CNV) is a highly prevalent medical condition which impairs visual acuity. The role of specific proteins in modulating CNV has been extensively reported, although no studies have described the entire human proteome in CNV corneas. In this paper, we performed a proteomic analysis of vascularized vs healthy corneal stroma, in a CNV mouse model and in CNV-affected patients, with a specific focus on extracellular matrix (ECM) proteins. We identified and quantified 2315 murine proteins, 691 human proteins and validated 5 proteins which are differentially expressed in vascularized samples and conserved in mice and humans: tenascin-C and fibronectin-1 were upregulated, while decorin, lumican and collagen-VI were downregulated in CNV samples. Interestingly, among CNV patients, those affected with Acanthamoeba keratitis showed the highest levels of fibronectin-1 and tenascin-C, suggesting a specific role of these two proteins in Acanthamoeba driven corneal CNV. On a broader picture, our findings support the hypothesis that the corneal stroma in CNV samples is disorganized and less compact. We are confident that the dissection of the human corneal proteome may shed new light on the complex pathophysiology of human CNV, and finally lead to improved treatments.


Murine cornea immunohistochemistry.
After microsurgical excision, the corneal stroma and epithelium were separated with 20 mM EDTA in phosphate buffer (PBS) for 30 min at 37 °C. Corneal stroma was then fixed for 15 min in ice-cold acetone, washed with PBS and blocked for 1 h with 2% bovine serum albumin (BSA; Sigma-Aldrich), 5% normal donkey serum (Sigma-Aldrich) in PBS. Tissue was incubated at 4 °C o/n with primary antibodies (rat anti-mouse CD31, BioLegend, and rabbit anti-mouse LYVE1, AbCam) in blocking solution, followed by secondary antibodies in PBS 2 h at RT (AlexaFluor 594 donkey anti-rat and AlexaFluor 488 donkey anti-rabbit, Invitrogen). After washing in PBS, corneas were radially cut and flat mounted on glass slides using the Vectashield mounting medium with 4′,6-Diamidino-2-Phenylindole (DAPI, Vector). Images of CNV were analyzed using an ImageJ-based system to quantify the total area of the cornea covered by vessels.
patients. CNV affected patients (n = 19, see Table 1) planning to undergo corneal transplant were selected, after obtaining informed consent and after approval by the San Raffaele Hospital Ethics Committee, by the Cornea and Ocular Surface Unit at the San Raffaele Scientific Institute. CNV was quantified by counting the number of corneal quadrants involved, with a score ranging from 1 (neovessels involving 1 quadrant) to 4 (neovessels involving four quadrants) at the time of corneal transplant. CNV patients in this study had an average CNV score of 3.05 ± 1.17. Fresh corneal buttons obtained from these patients immediately after keratoplasty were collected in Carry-C transport solution (Alchimia) for subsequent proteomics or Western Blot/ELISA analysis. Avascular, keratoconic corneas (n = 27, Table 1) were used as avascular controls. Methods used to obtain human data were performed in accordance with the relevant guidelines and regulations.
Murine and human cornea protein extraction. Murine pooled samples and human corneas were processed according to the following protocol.
Upon incubation in 20 mM EDTA with protease inhibitor cocktail (Sigma-Aldrich) for 40 min at 37 °C, epithelium was removed from the stroma using forceps with the help of a stereomicroscope. Corneal endothelium was not removed from the stroma, since it's impact on stroma ECM proteomics is negligible 26 . For the proteomic analysis, the stroma was cut in small pieces and put in 100 µl of R3 buffer (urea 5M, thiourea 2M, CHAPS 2% w/v, Zwittergent 2% w/v) plus protease inhibitor cocktail. Samples were subsequently homogenized with a plastic potter and incubated on a rotating shaker for 1 h at RT. After centrifugation (14,000 × g, for 10 min at RT), supernatants were used for proteome analysis.
Mass Spectrometry and proteomics analysis. Proteins were quantified using Bradford assay and BSA as standard. Proteins were digested using the FASP (Filter Aided Sample Preparation) protocol 27 . Briefly, 20 µg of total proteins from each sample were diluted to 200 µl with 0.1 M Tris/HCl pH 7.4. Cysteines were reduced with 45 µl of 0.1 M DTE (1,4-dithioerythritol) in Tris/HCl pH 7.4 and incubated at 95 °C for 5 min; then samples were centrifuged on Amicon Ultra-10K at 14,000 × g for 10 min with a solution of urea 8 M. Cysteines were alkylated with 100 µl 0.05 M IAA (Iodoacetamide) in urea 8 M added on the filter and laid at RT for 5 min; samples were centrifuged on Amicon Ultra-10K at 14,000 × g for 10 min with a solution of urea 8 M and then incubated overnight at 37  allowed; carbamidomethylation of cysteine as fixed modification; N-terminus-acetylation and Methionine oxidation as variable modifications; ±5 ppm and ±20 ppm for precursor and fragment ions mass tolerance, respectively. The proteins identified in murine and human stroma were subsequently compared with the ECM protein database Matrisome 29 to confirm the prevalent presence of stromal proteins in the analysed samples.
Human stromal samples were lysed as above reported for the murine samples. The only difference was the digestion with dextranase from Penicillum sp. 1 U/µl (Sigma-Aldrich) in 0.05 M KH2PO4 buffer pH 6 at 37 °C overnight prior to the Bradford assay. In details, human stromal proteins were incubated with dextranase in 1:4.5 (v/v) in order to hydrolyze the dextran, which derived from the medium used for cornea conservation. Proteins were subsequently digested using the FASP protocol and analyzed by mass spectrometry. Data Normalization and differential expression assessment. Protein identifiers were converted to Entrez Gene identifiers, and Homologene 30 was used to pair mouse and human proteins by homology. The mapping yielded proteins with a homologue available in both species and for which we detected valid (not null) proteomic measurements in all samples. Proteomics measurements were normalized using weighted-trimmed mean of M-values 31 and differential expression was assessed by means of the moderated t test 32 , two approaches initially proposed for gene expression data, but shown to be useful also for proteomics data 33,34 . Nominal p values were corrected using the BH procedure. The R programming environment 35 and its packages limma 36 and edgeR 37 were used to carry out normalization, differential expression and multiple testing correction. Proteins were considered differentially expressed at absolute values of log2 Fold Change (LFC) greater than 0.58, corresponding to FC less than 2/3 or greater than 3/2, and False Discovery Rate (FDR) less than 0.2. eLiSA. Human corneal stromas were cut in small pieces and resuspended in 100 µl PBS with protease inhibitor cocktail (Sigma-Aldrich). Samples were homogenized with a T110 homogenizer (IKA), 30 sec (power 5, 3 times on ice) and centrifuged at 12,000 × g for 10 min to remove tissue debris. Supernatants were quantified with Bradford protein assay (Thermo scientific, Waltham, Massachusetts, USA and 5 µg of total proteins were analyzed in triplicate with TNC and FN1 Elisa KITs (Cayman), following manufacturer's instructions.
Western blot. Five μg of proteins from control or CNV stromas were re-suspended in NuPAGE LDS reducing Sample Buffer (Thermo Fisher), resolved on NuPAGE 4-12% Bis-Tris Protein Gels (Thermo Fisher) and electro-transferred to nitrocellulose membranes (Amersham, Little Chalfont, UK) for Western blot (WB) analysis. Protein transfer was evaluated by red Ponceau S staining (Sigma-Aldrich). Membranes were blocked in a Tris buffered solution (TBS) 5% milk, 0.1% Tween 20 and incubated overnight with primary antibodies: mouse anti-human decorin (MAB143), rabbit anti-human lumican and anti-human collagen-VI α1 (NBP1-87726 and NB120-6588, Novus Biologicals) at 4 °C under gentle shaking. Subsequently, membranes were incubated at RT for 1 h with anti-mouse or anti-rabbit HRP-conjugated secondary antibodies (NA9310V and NA9340V respectively, Ge Healthcare) followed by chemiluminescence reaction performed with ECL detection reagent (Ge Healthcare) and film exposure. The protein band optical density was finally measured with the UVITEC imaging system. Expression of β-actin revealed with an HRP conjugated mouse monoclonal antibody (ab49900, Abcam) was used as loading control.
Statistics. Corneal vascularization rate over time was analyzed by One-Way ANOVA analysis, following Tukey multiple comparison tests. Unpaired t-test was used to evaluate the differences in WB band intensity and ELISA values between control and CNV patients. A p value < 0.05 was considered to be statistically significant. The statistical software GraphPad Prism 5.0 (GraphPad) was used for all analyses. All methods were performed in accordance with the relevant guidelines and regulations.
Murine and human cnV proteomics. Mass spectrometry (MS) analysis of murine corneal stroma throughout the entire vascularization period allowed us to identify and quantify a total number of 2315 proteins. We excluded the 180 days endpoint from our dataset for the proteomic analysis, because the number of the identified proteins was particularly small compared to the other time points. This was probably the consequence of a technical problem occurred during sample preparation. In any case, we did not see any significant difference of CNV extension at this time point by means of immunofluorescence. In human samples (8 controls vs 10 CNV), 691 proteins were detected. Interrogation of the ECM protein-database Matrisome retrieved 115 matrisome-associated proteins (Supplemental Fig. S1A www.nature.com/scientificreports www.nature.com/scientificreports/ core matrisome of ECM, 103 proteins were found, of which 66 belong to ECM glycoproteins, 25 to the collagen family and 12 to ECM proteoglycans (Supplemental Fig. S1B, Supplemental Table S1). In human samples, the comparison identified 74 matrisome-associated proteins with 43 ECM regulator proteins, 14 ECM affiliated proteins and 17 secreted factors (Supplemental Tables S1C, S2); out of the 93 proteins identified as part of the human core matrisome, 60 were ECM glycoproteins, 21 collagens and 12 ECM proteoglycans (Supplemental Fig. S1D, Supplemental Table S2). This data indicates a good coverage of the stroma proteome, both in murine (20%) and human (16%) samples, as expected when compared to the entire species-specific matrisome. Moreover, we could confirm the good technical extraction of ECM proteins by our tissue protein extraction method.
comparative analysis of murine and human cnV proteomics. The comparison of differential expression in human and mouse highlighted several proteins with significant expression changes in the same direction (cases vs controls) (Fig. 3A, green dots). Protein-protein interaction network of DEGs was constructed using the Search Tool for the Retrieval of Interacting Genes (STRING, http://string.embl.de/) database 38 . This procedure allowed us to identify 5 ECM proteins in common between mouse and human (STRING network, Fig. 3B), which are strongly related, often cited together in the literature, which interact biologically each other or which are part of the same family. Two of these proteins were upregulated in the murine sutured and human  Human protein validation. The differential expression of TNC, FN1, DCN, LUM and COLVI in the human corneal stroma was finally validated by WB or ELISA analysis. As shown in Fig. 4A,B, tenascin-C and fibronectin-1 levels were increased in CNV patients (+319% p = 0.0009, +102% p = 0.0047 respectively). On the other side, the expression of decorin (Fig. 4C), lumican (Fig. 4D) and collagen-VI α1 (Fig. 4E) was reduced (−85.8% p = 0.0018, −55.1% p = 0.002 and −59.1%, p = 0.0044 respectively). These data confirm the results obtained from the MS analysis of the human corneal stroma. Representative WB images are shown in Fig. 4F. For each validated protein, a graph showing the trend over time in the mouse model is reported on the right side of the corresponding panel.

Discussion
Corneal neovascularization is a leading cause of blindness worldwide and the need for better treatment is real and urgent. To this end, the quantification of the entire human CNV proteome is a useful tool to provide novel insights into the pathophysiology of this disease, and generate novel therapeutic targets.   Table 2. ECM proteins with a common expression changes in mouse and human. Values of log 2 fold change (LFC), p-value (p) and adjusted p-value (FDR) for each of the proteins with significant expression changes in the same direction (cases vs controls). Human and mouse data are reported (left and right columns respectively). www.nature.com/scientificreports www.nature.com/scientificreports/ Until today, a few studies focused on the physiological corneal proteome 26,39-41 , but none of them analysed the pathological proteome of human CNV, as we did here.
In order to improve the purity and optimize ECM protein retrieval in CNV mice and humans, we took action to reduce variability. Specifically, in order to avoid unwanted dilution of stromal ECM structural proteins, and differently from previous studies 13,14 , we removed the corneal epithelium prior to tissue processing. In addition, our suture-induced CNV model allowed us to follow-up animals for a long time (10 months), with less side-effects (corneal opacity, infections and perforations) compared to other damage models used in previous studies 13,14 .
We acknowledge that pooling together murine samples can reduce variability and, hence, be a limitation of this study, although this was required for ethical reasons to limit the number of sacrificed animals. On the other hand, however, we used completely independent samples and techniques (ELISA and Western Blot) to validate our proteomics results. It should also be noted that pooling the samples makes the pool stronger in terms of stability, because the inter-individual variability is attenuated.
Cursiefen et al. quantified the extent of CNV growth and regression using the same murine model we used 42 , showing long-term persistence of blood vessels (up to 8 months) and total regression of lymphangiogenesis after 6 months, in the murine cornea. We were able to observe corneal blood and lymphatic vessels until 300 days after suture removal. Nonetheless, our data confirm that vessel regression continues over time, with an initial one-week, fast regression followed by a second, slower regression phase. It should be noted, however, that CNV was quantified differently in the two studies and this may have influenced the results. In fact, we performed whole-mount staining and quantified the vascular area on the entire cornea, while Cursiefen et al. performed quantification on corneal cross sections.
Our novel approach of CNV proteome analysis at different time points during the entire 10-month period increases the relevance of the animal model in terms of clinical translation, since patients affected with long-standing CNV are by far more numerous.
The proteomic analysis allowed us to identify, for the first time, five closely related ECM proteins, which are differentially expressed in vascularized human corneas.
Specifically, we found that two proteins: tenascin-C and fibronectin-1 were upregulated in the CNV cornea. Tenascin-C is highly conserved in vertebrates 43 , mostly expressed by stromal cells and leukocytes at sites of inflammation, including the cornea after refractive surgery [43][44][45][46][47] . www.nature.com/scientificreports www.nature.com/scientificreports/ Interestingly, the increased expression of tenascin-C is maintained not only in acute, but also in chronic inflammation, where it contributes to disease severity 48 , which explains its presence in the setting of long standing CNV in our study. Tenascin-C promotes vascularization by favoring cell spreading and signaling 49,50 and by reducing tissue rigidity 51 , which in turns promotes vascular endothelial cell invasion. Tenascin-C targeting therapies with RNA interference have been already proposed 52,53 , even in very recent studies 54 for the treatment of glioblastoma, impairing the tumor cell migration and invasion. Our findings support a potential use of a similar TNC inhibitory approach for CNV treatment.
Fibronectin-1 is a well-known ECM protein that mediates cell adhesion, migration, growth and differentiation 55 . Of note, TNC and FN1 are expressed during the early phases of wound healing, when neovascularization is more intense 56 . Intriguingly, TNC binds to FN1 57 and seems to regulate the pro angiogenic effects of FN1 in tumors 58 . FN1, on the other side, binds to and regulates decorin function 59 .
Infectious keratitis is a leading cause of CNV. In our study, 3 patients with CNV following Acanthamoeba infection were analyzed and we surprisingly found that they showed the highest concentration of FN1 and TNC (+116%, p = 0.044 + 49.5% p = 0.012 compared to non-Acanthamoeba CNV patients). In fact, Acanthamoeba, can bind to ocular FN1, and this is instrumental to the development of keratitis 60 . The potential role of FN1 overexpression in inducing CNV is supported by the finding that FN1 concentration is increased in contact lens wearers 61 , which are both at higher risk of developing CNV 62 and Acanthamoeba keratitis 63 . We are aware that drawing conclusions from a population of only 3 patients is speculative: confirmation of our findings in a larger cohort of Acanthamoeba keratitis patients will be necessary, in the future, to confirm these findings.
FN1 is produced by pro-angiogenic macrophages 64 and stimulates endothelial cell proliferation, micro-vessel elongation and angiogenesis 65 . Finally, and in line with our observation of extensive lymphangiogenesis in mice, FN1 is an excellent substrate for lymphatic endothelial cells invasion 66 , while TNC induces lymphangiogenesis in tumors 67 . Interestingly, several studies have targeted FN1 over-expression in tumors, with therapeutic molecules or even imaging agents 68,69 . This approach can be now proposed in CNV, thereby achieving in situ and highly-specific vascular inhibition at the site of vessel growth.
On a different note, we found that decorin and lumican were downregulated in vascularized mouse and human corneas. These proteins belong to the small leucine-rich proteoglycan family (SLRP) of the ECM 70 . They are expressed in the normal corneal stroma, where they promote corneal transparency, collagen fibril assembly and increase tissue rigidity by stabilizing collagen architecture 71 . Previous reports support the role of decorin in acute (up to 14 days) CNV in rabbit 72 . Further, it is known that deletion or mutations of DCN and LUM can lead to corneal dystrophies [73][74][75] . However, the relevance of decorin in human and in non-acute CNV has not been tested before.
It is known that DCN contributes to the maintenance of proper collagen organization 8,76 and increases corneal stiffness 77 . The fact that DCN expression is reduced in CNV supports the concept that decreased tissue rigidity is associated with vascular invasion in the cornea, as it occurs in other tissues 24,78 . In this vein, it has been shown that increasing corneal rigidity by corneal crosslinking can inhibit CNV 79 . In addition to its effects on tissue biomechanics, overexpression of DCN retards in rabbits CNV, by down-regulating pro-angiogenic VEGF and angiopoietin 72 . In tumors, DCN-based therapies can promote tissue regeneration and reduce fibrotic scarring 80 , by downregulating several growth factors 81 .
Similarly to DCN, LUM also promotes collagen deposition, organization 82 and stabilization 83 , therefore promoting tissue rigidity, and hence, a-vascularity. Moreover, lumican inhibits angiogenesis by interfering with α2β1 activity and downregulating MMPs expression 84 . In addition, it inhibits the influx of inflammatory cells to the cornea [85][86][87] , which is instrumental to CNV development 88 . Our finding confirms the results of a recent study 89 , which showed that LUM derived peptides promote corneal wound healing; this opens up a new promising field for the treatment of corneal scarring and CNV.
Finally, we observed reduced expression of collagen-VI, which is constitutively expressed in the corneal stroma 90 . Reduction in COLVI could promote CNV by altering the corneal biomechanics. For instance, COLVI depletion in the adipose tissue and tendons results in disorganized tissue morphology 91,92 . Furthermore, very recent studies reported that the C-terminal product of collagen-VI α-3, endotrophin, promotes angiogenesis and inflammation through recruitment of macrophages and endothelial cells in tumors 93 .
In addition to the five proteins whose differential expression is conserved in mouse and human subjects, we provide a list of other proteins, which are differentially expressed in mice or in human CNV samples, which will be subject of future studies.
In summary, our data globally suggest that the corneal stroma in CNV patients is highly disorganized and less rigid. These morphological modifications follow the release of soluble factors from leukocytes 94,95 , endothelial cell invasion and secretion 96 -which also contribute to tissue rearrangement-and vascular leakage/edema, with lymphatic vessel infiltration 97 .
In conclusion, in this paper we identify, for the first time, five ECM proteins, which are conserved in humans and are differentially expressed in normal vs. vascularized corneas. We suggest that our findings may shed new light on the complex pathophysiology of human CNV, and finally lead to improved treatments.